While going over sparkling-water examples, a common pattern that is seen is for scoring and collection scores over a h2oframe is to do the following:
val predictionH2OFrame = dlModel.score(result)('predict)
val predictionsFromModel = asRDD[DoubleHolder](predictionH2OFrame).collect.map ( _.result.getOrElse("NaN") )
I need to understand if the order of scores in the original frame will be maintained using this approach. In particular, will the scores order be maintained through the the cast to a RDD followed with the collect?
Yes, H2O guarantee the order. You can safely join input frame with prediction frame.